Skip to content

Loopy model

Drive portrait motion from audio alone

Visit official site

What Loopy is

The model is described as an end-to-end audio-driven video diffusion model with cross-clip and intra-clip temporal modules and an audio-to-latent module. It uses long-term motion information from data to learn natural motion patterns and removes the need for manually specified spatial motion templates.

What you can do with it

  • Generate vivid motion details from audio across visual and audio styles
  • Produce motion-adapted results for one reference image under different audio
  • Support non-verbal actions such as sighs, brow and eye movements and natural head motion
  • Handle fast, gentle or lifelike singing and accept side-profile images

Application scenarios

  • Create speech-synced avatars for video meetings
  • Give game characters responsive facial expressions in virtual reality
  • Make personalised animated portraits for social platforms

Getting started

  • Visit the official site or the GitHub page
  • Read the documentation on how the model works and its conditions of use
  • Download the required code and datasets
  • Set up the environment with the needed libraries and dependencies

Pros & cons

✓ What we like

  • Non-verbal actions such as sighs, brow and eye movements are supported
  • Side-profile images are accepted as input

! What to watch out for

  • Motion is generated from audio input and a reference portrait
  • Setup requires downloading code and datasets and installing libraries

FAQ

What drives the motion?

The material describes an end-to-end audio-driven video diffusion model.

What inputs are supported?

The feature list names different audio inputs, a reference image and images with side profiles.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review